AI Disclosure Penalty: When the Label Matters More Than the Text

A 16-experiment study shows that declaring AI use can reduce the perceived authenticity and value of creative writing. A GDE reading of experience, trust, responsibility and the edge of the known.

Andrea Viliotti

AI disclosure penalty

There is a paradox that every leader, author, consultant and company using generative AI will need to understand. We can write faster, explore more sources, organise more alternatives and produce more polished drafts. Yet the moment we tell the reader that AI was involved, the perceived value of the text may fall.

The point is not simply that the text becomes worse. The paper by Manav Raj, Justin M. Berg and Rob Seamans is more subtle: the label attached to the text can change the reader’s judgement before the reader has fully encountered the text itself.

For a company, this is not an academic detail. It affects how a brand communicates, how a founder writes to stakeholders, how a consulting firm presents an argument, how a knowledge team publishes insight and how any organisation makes transparent use of AI without weakening trust.

In brief

The AI disclosure penalty is the reduction in perceived value that can emerge when readers know, or believe, that a text was created by AI or with AI assistance.

The paper connects this penalty primarily to perceived authenticity: the text is experienced as less crossed by human experience.

The GDE reading proposed here shifts the managerial question. The issue is not simply “AI or no AI?”, but which human judgement, which sources and which responsibility make the text trustworthy.

What the paper actually measures

The paper “The Artificial Intelligence Disclosure Penalty: Humans Persistently Devalue AI-Generated Creative Writing”, published in the Journal of Experimental Psychology: General, examines a very practical problem: what happens when readers believe that a piece of creative writing was produced by AI, or with AI support?

Its strength lies in scale and experimental discipline. The authors report 16 preregistered experiments, with a total sample of 27,491 participants, conducted across a period in which generative AI rapidly entered public attention. Participants evaluated writing samples under different disclosure conditions.

The central finding is clear. When participants believed that a text was written by AI, or with the help of AI, they tended to evaluate it less favourably than when they believed it had been written by a human author without AI. In some designs, the same or comparable content was devalued because of the attributed origin.

The paper also identifies a central psychological mechanism: perceived authenticity. Readers do not merely discount the text because they assume it is technically weaker. They discount it because the AI label makes the text feel less authentic, less humanly grounded and therefore less worthy of appreciation.

Another important finding is persistence. The authors test different metrics, contexts, kinds of writing and interventions. They also examine ideas such as humanising the AI, presenting AI as a tool and framing the work as human–AI collaboration. The penalty proves difficult to moderate in a stable way.

It is equally important not to ask the paper to prove more than it proves. It does not show that AI always writes worse than humans. It does not show that every reader in every culture will respond in the same way. It studies creative writing, mainly in English, with online participant samples, during a specific period of AI diffusion. Its value is narrower and stronger: it shows that disclosure can change evaluation even before a text is judged only on its content.

The penalty is not only in the text; it is in the label

A GDE reading starts from a simple distinction: the reader never encounters only a text. The reader encounters a text plus a label of origin. “Written by a person” and “written with AI” are not neutral tags. They alter the observational relationship between reader and artefact.

This is the part most relevant to leaders and organisations. Many companies treat AI disclosure as a compliance note: “AI tools were used”. But a reader does not receive this sentence only as technical information. The reader may receive it as a signal about intention, experience, effort and responsibility.

The paper suggests that the threshold of attention can move. If the reader associates AI with weaker authenticity, the reader may lower the willingness to recognise value. Not because every sentence has been analysed in detail, but because the label has already changed the frame.

The disclosure penalty, therefore, is not only a content problem. It is an observer–artefact problem. The text is the artefact. The reader is the observer. The label “AI” modifies the relation between the two.

Authenticity, experience and memory artefacts

To understand why authenticity matters so much, we need to distinguish direct experience from indirect experience.

Direct experience is what we have lived through. It is not only what we can describe; it is what has changed the way we see. A founder who has survived a liquidity crisis understands a sentence about cash discipline differently from someone who has only read a finance manual. The same words carry a different weight because they are connected to lived experience.

Indirect experience is the echo of someone else’s experience. We read a book, a research paper, a novel, an article, and we receive the symbolic transformation of something another person has lived, studied or imagined. We do not directly absorb that experience. We receive its echo, and that echo becomes meaningful when it resonates with our own experience.

From this perspective, every cultural artefact is a device of memory. A text, an image, a theory, a sculpture, a business procedure: each translates experience into a form that can circulate beyond the person who created it.

The paper shows that when readers believe AI is behind the text, this perception of authenticity may weaken. The GDE interpretation is that the AI label can break, or at least disturb, the expected chain between text and human experience. The reader starts to wonder: who has lived, selected, judged and risked something behind these words?

AI as the echo of an echo

Generative AI should not be reduced to a cheap statistical trick, but it should not be transformed into a conscious author either. It works on symbolic traces: texts, images, data, code, fragments and patterns left by human activity and by previous systems of representation.

For this reason, with a controlled metaphor, generative AI can be described as the echo of an echo. Human experience becomes language, data, images and archives. AI crosses those archives, detects regularities, recombines patterns and produces a new surface. That surface can be useful, surprising and even beautiful. But its relation to experience is indirect twice over.

Here lies the contradiction. Many human texts are not pure direct experience either. Human authors also write through books, teachers, traditions, memories, citations, methods and borrowed structures. The difference is not that the human text is always original and the AI-assisted text is always derivative. The difference is whether the reader can perceive a responsible human criterion behind the final form.

The decisive question is not whether AI is allowed or forbidden. The question is where human judgement is located. Who chose the direction? Who separated relevance from noise? Who checked sources? Who accepted responsibility for the interpretation? Who transformed a generated surface into a meaningful act?

Without this clarification, AI disclosure may feel like a subtraction of experience. With this clarification, it can become a more mature form of transparency.

The blind spot of the human reader

The paper is important because it reveals not only a possible limit of AI, but also a limit of the reader. The reader believes he or she is evaluating the text, while also evaluating the label that precedes the text.

This is understandable. The human mind cannot explore everything with equal intensity. It uses shortcuts. It searches for signals of reliability, proximity, intention, effort and shared experience. Human origin is one such signal. It tells the reader that there may be a body, a history and an experience behind the words.

A useful shortcut, however, can become a blind spot. If every AI-assisted text is downgraded before it is read, part of the field disappears. Not because AI is automatically profound, but because some AI-assisted work may contain genuine human judgement, real sources, careful interpretation and a meaningful contribution.

This is where the figure of the explorer enters. The standard reader tends to remain near the centre of what is already recognised. The explorer moves toward the edge: the place where the known becomes unstable, where a new question appears, where a pattern has not yet become common language.

Exploring the known is already a vast task. No company, author or researcher can cross the entire symbolic archive of humanity alone. AI can help map that archive, discover associations, compare arguments and reveal tensions. It does not replace the explorer. It can extend the explorer’s reach inside the known.

Zero, prime number, infinity: a GDE grammar of novelty

The GDE trace uses three mathematical images: zero, prime number and infinity. They should not be read as mathematical proofs of a psychological phenomenon. They are structural analogies, useful for thinking about novelty, observation and the limits of AI.

Zero is not merely absence. It is the initial point of observation. It is the place from which a sequence becomes readable. Without a zero point, we do not know where we are looking from, where a path begins or how a movement can be oriented.

In this analogy, the prime number is the observable emergence of a novelty that cannot be reduced in a banal way to the recombination of previous elements. Of course, a prime number belongs to the sequence. It does not fall from outside the system. Yet it appears as something that resists simple decomposition. It is not just another rearrangement of what came before.

Infinity is the edge. It is not an object we possess, but the operational limit of the system. To measure it completely would transform the system itself. Every organisation knows this in practical terms: there are limits that cannot be crossed without changing the identity, resources or structure of the organisation that tries to cross them.

Where does AI sit in this grammar? AI is powerful inside the sequence of what has already been observed and symbolised. It can move across numbers, texts, cases, patterns, analogies and sources. It can show the density of the known better than any individual human observer.

But it does not own the next prime number before that number has emerged. It does not experience the new before it becomes observable. It can, however, be useful near the edge. Not because it can say with certainty what will emerge, when it will emerge or who will carry it. It can signal pressure: areas where the known no longer closes, where contradictions accumulate, where a new observer may be needed.

What leaders and companies should understand

For a company, the lesson is not “hide AI”. That would be a poor and risky conclusion. The lesson is to design a better pact of authenticity.

When an organisation discloses AI use, it should clarify both the role of AI and the role of the human author or team. A sentence such as “content generated with AI” is too broad. It can mean full automation, assisted drafting, source exploration, translation, editing, summarisation or a deep human–AI research process. The reader cannot know which one unless the organisation explains it.

A useful disclosure should answer five questions. What part of the work was assisted by AI? What part was decided by a person? Which sources or data were used? Which interpretation remains the responsibility of the author or organisation? What value has been added beyond the generated surface?

These questions matter for marketing, institutional communication, consulting, training, knowledge management and strategic writing. A generic AI-produced commercial text may weaken trust. A well-governed AI-assisted text, where human criterion is visible, can strengthen the reader’s understanding of the work behind the words.

For brands, this will become increasingly important. The volume of content will continue to grow. The problem will no longer be producing correct sentences. The problem will be demonstrating why those sentences deserve attention.

Disclosure: from formal note to responsibility pact

AI disclosure can become a defensive formula or a new grammar of responsibility. The difference lies in how it is designed.

The defensive formula says: “An artificial intelligence tool was used.” It may protect formally, but it leaves the reader with a gap. Who is speaking? Who is responsible? Which part of the text comes from experience and which part from automation?

A responsibility pact says something different: “This text was built with AI support for source exploration, alternative generation and structural drafting. The judgement, selection, interpretation and final responsibility remain human.” This formulation does not remove all risk, but it restores the reader’s path to the authorial centre.

This matters especially when the text is creative, strategic or reputational. A technical manual can be evaluated largely on correctness. A letter to employees, a founder’s note, a strategic essay or a brand manifesto requires more: it requires an intelligible chain of experience.

The managerial question is no longer “AI or no AI?”. It is: which experience, which judgement and which responsibility have crossed this text? Where the answer is weak, disclosure becomes a liability. Where the answer is strong, disclosure becomes an act of trust.

Conclusion — AI does not erase the author

The paper by Raj, Berg and Seamans does not prove GDE, nor does it claim to explain the entire relationship between humans and artificial intelligence. It measures a specific and important phenomenon: the AI label can reduce perceived value, largely because it weakens perceived authenticity.

The GDE reading suggests that the issue is not only technological. It is observational. The reader looks for a signal of embodied experience in the artefact. When the reader sees the AI label, that signal may become weaker, or more ambiguous.

This does not mean that the author disappears. It means that authorship changes. The author is no longer only the person who writes every word. The author is the person who defines the zero point of observation, recognises where the known is not enough, decides which echoes deserve to become experience, and accepts responsibility for the final form.

For companies, this is the real challenge. Not to produce more content, but to produce content in which the chain of experience can still be read. In the world of generative AI, trust will not come from hiding the tool. It will come from making human judgement visible.

Key questions for anyone publishing AI-assisted content

What is the AI disclosure penalty?

It is the reduction in appreciation or evaluation that can appear when readers know, or believe, that a text was produced by AI or with AI support. In the paper analysed here, the central issue is not only textual quality but the authenticity readers attribute to the work.

Why is perceived authenticity so important?

Because readers evaluate more than words, style and structure. They also look for signals of experience, intention and responsibility. When the AI label weakens those signals, the content may be devalued before it is fully considered.

Should companies hide AI use?

No. The stronger response is not concealment, but better disclosure. Organisations should explain what AI did, what humans decided, which sources were used and what distinctive value was added by human judgement.

What is the right question for leaders, authors and consultants?

Not simply “AI or no AI?”, but: which experience, which judgement and which responsibility have crossed this text? When that chain is visible, disclosure can become a pact of trust rather than a defensive note.